In distributed computing systems, images and video captured by devices at the edge of a network are typically compressed and transmitted to the cloud for further processing. The images and video must subsequently be decompressed before any processing can be performed. This approach is inefficient, however, as it often wastes significant bandwidth on transmission of insignificant and/or trivial visual data, it requires full decompression as a pre-requisite to performing any meaningful processing and analysis on visual data, and it fails to leverage the processing capabilities of other nodes in the network efficiently.